Value at Risk and Market Risk

Eric Falkenstein
 VaR: The Big Picture

• VaR is the new standard for measuring market risk across

a Variety of security types

• VaR should always be complimented with specific

scenario tests

• VaR is integral in any attempt to allocate capital (for both

economic and proposed regulatory charges)

2 Eric Falkenstein 4/14/99

 Subtle Benefits of a VAR System

• Necessitates the ability to mark a book to market, which by

itself is a major preventative against unpleasant surprises
• Establishes a clear independent risk monitoring arm
• Helps with existing auditing and risk reviews
• Securities that are not understood well and therefore have a
significant risk are flagged as exceptions more often
• If used for capital charge, gives an incentive to economize
on risk even when risk limits are not breached

3 Eric Falkenstein 4/14/99

 What Can We Learn From These?
Worst-Case Derivative Scenarios

First Reported Company Estimated loss (millions)

First Reported Company Estimated loss (millions)
11/92 Dell $8 10/94 Community A Mgmt. $44
2/93 Showa Shell $1,700 10/94 Portage County OH $8
7/93 Nippon Steel $128 10/94 Sears $237
1/94 Metallgesellschaft $1,300 11/94 Todyo Securities Co. $356
1/94 Codelco $200 11/94 3 Farm Credit System Banks $23
3/94 Askin Capital Management $600 12/94 Orange County CA $1,700
3/94 Minnetonka Fund/Cargill $100 12/94 Chemica Bank $70
4/94 Mead Corp $7 2/95 Barings $1,200
4/94 Procter & Gamble $157 3/95 Winsconsin Inv. Board $95
4/94 Gibson Greetings $23 3/95 MCN Corp $10
5/94 Air Products $69 5/95 Five Morgan Stanley clients $28
5/94 ARCO $22 7/95 First Capital Strategists $137
6/94 Dell $35 8/95 Postipankki (Finland) $110
6/94 Virginia Retirement System $66 10/95 Daiwa $1,100
6/94 Florida Treasurer's Office $175
6/94 Pain Webber Bond Mutual Fund$33
7/94 Glaxo Holdings $100
7/94 CS First Boston Inv't Mgt. $40
8/94 Piper Jaffray $700
8/94 Charles County $2
8/94 Colonia (Germany) $76
9/94 Shoshone Indians $5
9/94 Investors Equity Life $90
9/94 Odessa College TX $11
4 Eric Falkenstein 4/14/99
 VaR is the Best Market Risk Metric

• Do not compare VaR to a Platonic ideal, but to feasible

alternatives

• Most importantly, by measuring VaR more precisely, the

right questions get asked which minimizes “operating risk”

• While VaR won’t catch the next Baring’s, one can also bet
that the next Baring’s won’t have a fully operational VaR
risk measurement system in place

5 Eric Falkenstein 4/14/99

 What Did We Have Before VaR?

Duration
• Duration has a practical application in the measurement of
a bond’s risk
% Change in Price = Duration x Change in Yield
• Given a parallel shift in yields, we can know how different
securities will react
example:
• A bond with a duration of 5 years will experience a 0.35%
change in price if rates rise 7 basis points (5*.07)

6 Eric Falkenstein 4/14/99

 Problems with Duration

• Duration measures the risk from parallel shifts in rates;

rates do not always move in parallel

• Not applicable to FX

• Not informative for securities with significant option

characteristics (e.g., mortgages, caps/floors, callable
bonds)

• Does not capture risk from changes in credit spreads

7 Eric Falkenstein 4/14/99

 Problem 2: How do you Measure the
Risk of This Portfolio?

Security Notional
Bonds $ 50
Interest Rate Swaps $100
Interest Rate Futures Contracts $150
Caps/Floors $ 45
FX Forwards $130
Total $475

Aggregate Notional is not meaningful in this context

8 Eric Falkenstein 4/14/99

 Problem 3: Judgement

Expert systems: we should have someone knowledgeable at

the helm who can “know” what is happening.

Problem: how do you aggregate a bunch of expert opinions?

How do you validate them?

9 Eric Falkenstein 4/14/99

 VaR Can Give a Meaningful
Aggregate Number

Security Notional VaR

Bonds $100 $ 3
Interest Rate Swaps $100 $ 5
Interest Rate Futures Contracts $100 $ 4
Caps/Floors $100 $ 4
FX Forwards $100 $ 7
Total $11

10 Eric Falkenstein 4/14/99

 So What is VaR?

• Assume you had a portfolio that consisted, trivially, of the

30 year Tbond. Your daily P/L would look like this:
Daily Profit and Loss on a 30 yr Tbond
3

2
Percent Change

95% VAR

-1
99% VAR

-2

-3
1995

11 Eric Falkenstein 4/14/99

 So What is VaR?

• Converting this information into a histogram, we get:

Daily Profit and Loss on a 30 yr Bond, 1995
60

50
Frequency of Occurances

40

30
95% VAR

20 99% VAR

10

0
-2.2 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
Percent Change Buckets

12 Eric Falkenstein 4/14/99

 In Words

• VaR is the expected least amount lost 5% of the time

or,
• VaR is the expected most amount lost 95% of the time
• In both of the above examples, you could use a 99% VaR
or any other number, and also use different time horizons
(e.g., 95% lost over daily observations, or annual
observations)
• VaR takes into account current estimates of the volatility of
various fundamental factors (e.g., on-the-run US Treasury
rates) and their covariances (especially important when
looking at a hedged portfolios)

13 Eric Falkenstein 4/14/99

 VaR: One Factor Example

• $1MM 30yr Bond’s DV01=$1400

• Standard Deviation of 30yr bond=4.5 b.p.s

 $change 
95%VaR # of stdev' s  risk factor stdev  delta 
 unit change in factor 
$1400
95%VaR  1.65  4.5bp   $10,510
1bp

14 Eric Falkenstein 4/14/99

 VaR: 2 Factors

• Portfolio is $1MM 30yr bond

5yr Key Rate DV01=$100
30yr Key Rate DV01=$1300
 5 yr  5bp,  30 yr  6bp, 10 yr ,5 yy  .94
95%VaR  vector of port. sensitivit y to factors correl. of factors vector of ...
 $100 $1300  1 .94  1.65  5 100 
95%VaR  1.65  5bp  1.65  6bp    
 bp bp  .94 1  1.65  6  1400 
 1 .94  825 
95%VaR  825 12870  
 .94 1  12870 
95%VaR  $13,648

15 Eric Falkenstein 4/14/99

 In General

• Calculate all portfolio sensitivities to material risk factors at 95% level

• Aggregate them taking into account correlations
VaR  S S
S=an nx1 vector of a portfolio’s change in value to a 95% worst-case scenario
change in the n underlying factors
=correlation matrix
example:
• Rate Risk for Bond Option Trading=$86,216
• Volatility Risk for Bond Option Trading=$48,406
• Assume the correlation of those risks is zero
1 0  86216 
95%VaR  86216 48406  
0 1  48406 
95%VaR  $98,875

16 Eric Falkenstein 4/14/99

 Revaluation Approach for
Nonlinear Instruments
• Replace 1.65 with the calculated  in the value of the derivative for
a 1.65 change in the underlying risk factor:

17 Eric Falkenstein 4/14/99

 Further Refinements

• Which volatilities to use?

Examples: moving average, exponential moving average,
GARCH, and implied volatilities from options
• If a Revaluation approach--Monte Carlo or Historical?
• Do you incorporate spread shocks (e.g., Fed Funds-Libor
spread)?
• Do you incorporate volatility shocks (e.g., cap vols rise)?

Always weigh benefits versus costs of implementation

18 Eric Falkenstein 4/14/99

 Criticisms of VaR

• If refinements are not addressed, VaR could be a poor

representation of risk (e.g., using delta-normal VaR for a
book with significant spread risk and optionality)

• Solutions are within the VaR paradigm

• Any good tool needs to be used wisely to add value

19 Eric Falkenstein 4/14/99

 Always Compliment with Stress Tests

VaR Up 35 bp Down 35 bp
Portfolio $6.0M -$26.2M $25.8M

• VaR tells you how much you lose with a known

probability, but with an unknown scenario

• Stress tests tell you how much you lose in a known

scenario, but with an unknown probability

20 Eric Falkenstein 4/14/99

 Combine VaR with Stress Tests to
get the Complete Picture

VaR

Gives best estimate of Robust Risk Measurement

worst-case P&L

Stress Tests

1) User defined scenarios

2) Shows where portfolio loses money
21 3) Anticipates “non-normal” events Eric Falkenstein 4/14/99
 One Should Always be Checking
Various Assumptions

Usually VaR measures risk on an end-of-day portfolio.

Using historical P&Ls, we can refine VaR estimates to
adequately capture:

• Intraday trading
• Peculiar nonlinearities and spreads
• Particular liquidity of the trader portfolio

22 Eric Falkenstein 4/14/99

 Validation is Key

• Validation is required for use of VaR as a measure of

capital

• Validation also helps one understand the relative

importance of various VaR refinements

• Each portfolio or trader will have different relevant risk

factors and pricing models, and it is the risk manager’s job
to find and quantify these risks

23 Eric Falkenstein 4/14/99

 Backtesting
BACKTESTING TYPE 1
A test of the validity of the simulation method.

A daily comparison of:

EX-ANTE VAR vs. EX-POST HYPOTHETICAL P & L

BACKTESTING TYPE 2
A test of the “goodness” of VAR as a predictor of worst case P/L loss in trading
A daily comparison of:

EX-ANTE VAR 99% Confidence Level vs. EX-POST ACTUAL P&L

Under BIS rules,

Multiplicative factor (“K”) depends on number of excesses in one year, where excess means that
on a given day:

ACTUAL P/L LOSS > VAR 99% C.L.

24 Eric Falkenstein 4/14/99

 Regulatory Capital

APPROACHES FOR
DETERMINING
REGULATORY CAPITAL

RULE BASED INTERNAL MODELS PRE-COMMITMENT

"STANDARD MODEL" With Standard Parameters APPROACH

Specific General Market

Risk Risk

25 Eric Falkenstein 4/14/99

 Regulatory Capital
FOR GENERAL MARKET RISK

- CAPITAL BIS = K * VAR BIS_GENERAL

K DEPENDS ON BACKTESTING

factor is the following:

Number of exceptions Factor
<=4 3
5 3.4
6 3.5
7 3.65
8 3.75
9 3.85
>=10 4
- PARAMETERS FOR VAR BIS

– OBSERVATION PERIOD FOR SIMULATION:

• AT LEAST ONE YEAR

– HOLDING PERIOD
• ONE DAY I.e., assume static portfolio

– CONFIDENCE LEVEL:
• 10-DAY “INSTANTANEOUS” SHOCK IN MARKET RATES
26 • 99% CONFIDENCE LEVEL Eric Falkenstein 4/14/99
 Regulatory Capital
FOR GENERAL MARKET RISK AND SPECIFIC MARKET RISK:

IF FULLY CAPTURE SPECIFIC RISK

Correlation Risk
Event Risk

THEN: REGULATORY CAPITAL BIS = K * VAR BIS_TOTAL

K DEPENDS ON BACKTESTING

IF ONLY CAPTURE CORRELATION RISK

THEN: REGULATORY CAPITAL BIS = K * VAR BIS_TOTAL + VAR BIS_SPECIFIC

K DEPENDS ON BACKTESTING

27 Eric Falkenstein 4/14/99

 VaR and Regulatory Capital

• The BIS guidelines suggest estimating capital by using 3

or 4 times the VaR plus Specific Risk

• Specific risk is either the Standardized amount or that from

internal models (not to be less than 50% of the
Standardized amount)

28 Eric Falkenstein 4/14/99

 The Standardized Regulatory Capital
Approach for Specific Risk Capital

• 8% against most assets

• 1.6% against OECD bank debt

• 0% against OECD government debt

• 4% against liquid equity portfolio

• 2% against equity indexes

• apply to both long and short positions

29 Eric Falkenstein 4/14/99

 Internal Models Calculation of
Specific Risk Capital

• If no specific credit risk measurement is operational,

4xMVAR

• If a model demonstrably isolates spread, event, and default

risk, 3xSVAR

• If a model can capture some but not all of specific risk,

4xSVAR

30 Eric Falkenstein 4/14/99

 Standardized Vs. Internal Models

• For individual security holdings, the standardized

regulatory approach allocates less capital for speculative
grade bonds and equities

• For portfolios, the internal models approach generates

lower capital requirements

• The BIS has therefore created an employment program for

quants

31 Eric Falkenstein 4/14/99

 Economic Risk Capital (Not Regulatory)

• Use 99% VaR over reasonable and flexible time-to-close assumption

• Take into account dynamic strategies in the form of loss limits

Think of an unused loan commitment:

One translates this into a Loan Equivalent Exposure, taking into

account the probability of being used in the future.
Example:
Trader A VaR Loss Limit
$45 $120

Capital=VaR+VaR Equivalent Unused Commitment

Capital=VaR+factor(Loss Limit - VaR)
Capital=$45+.5*(120-45)=$82.5

32 Eric Falkenstein 4/14/99

 Economic Risk Capital

A loss limit over the VaR is an unused commitment. Thus, one must
translate this into a VaR-equivalent exposure

Example:
Trader A 99% VaR Loss Limit
$45 $120

Capital=VaR+factor(Loss Limit - VaR)

Capital=$45+.5*(120-45)=$60

33 Eric Falkenstein 4/14/99

 Loss Limit Risk is Diversifiable Too

• How does one allocate while accounting for diversification?

Example:
VaR VaR-Equivalent Stand-Alone
Unused Commitment Capital

Trader A $42 $26 $68

Trader B $23 $13 $36
Total $50 ? ?

• We will suppose the total VaR-Equivalent Unused Commitment is reduced

similar to the VaR:
50/(42+23)=.77 Total VaR is 77% the sum of the parts

Stand-Alone Capital Allocated Capital

Trader A $68 $52(=68x.77)
Trader B $36 $28(=36x.77)
Total $80
34 Eric Falkenstein 4/14/99
 Important Notes

• Using VaR and Loss Limits and tying this to a capital

charge explicitly incents traders to minimize risk by both
lowering their daily VaR and loss limits

• Economic capital different than VaR used in regulatory

reporting

35 Eric Falkenstein 4/14/99

 Integrating Regulatory and
Economic Capital

• Case 1
Regulatory Capital = Economic Capital
No problem

• Case 2
Regulatory Capital < Economic Capital
Allocate Economic capital, since economic capital is the
binding constraint

• Case 3
Regulatory Capital > Economic Capital
Not clear
36 Eric Falkenstein 4/14/99
 Return on Equity

• Most successful traders generate exceedingly high ROEs.

• Nonetheless, ROE is a useful performance measure,

especially for new traders (e.g., writing calls on the S&P
would have been very profitable over many consecutive
months, yet on an ROE basis would have been very weak)

• P&L still reigns supreme in performance measurement,

and is essential for validation. Risk managers need to
know a trader’s P&L and their incentive compensation
plan.

37 Eric Falkenstein 4/14/99

 2 Ways that Bad Traders
are Exposed

• Blow out. A quiescent market can allow a negative NPV

strategy to produce seemingly large returns with low risk
for many consecutive months (e.g., Orange County)

• Profit drip. A trader might have very low risk, yet have
locked in a negative carry through poor pricing

In both cases, accurate monthly P&L helps

highlight these problems

38 Eric Falkenstein 4/14/99

 Asset and Liability VaR

• A Balance Sheet Management department has a different

risk profile than a trader, as trading horizons are typically 3
months to 1 year in duration

• Accounting measures of risk are the industry norm

• Deposits will not be marked-to-market anytime soon,

making VaR more abstract than accounting risk measures

39 Eric Falkenstein 4/14/99

 Balance Sheet VaR

• Like trading risk, capital should be allocated against the

current risk profile, plus a portion of the maximum feasible
risk exposure under corporate policy

• Often duration limits of 1 or 5 years act as the maximum

interest rate risk limit

40 Eric Falkenstein 4/14/99

 Incorporate Other Risk Factors

• 3 Factors of Yield Curve risk (shift, twist, curvature)

• Spread risk

• Prepayment risk

• Parameter uncertainty

41 Eric Falkenstein 4/14/99

 Benchmarks for BSM VaR

• Call Reports are not very informative

• Most banks have duration between 1 and 5 years

• When is a noisy VaR worse than an accounting measure of

Risk?

42 Eric Falkenstein 4/14/99

 Do Banks Have a Comparative
Advantage in Riding the Yield Curve?

• Average duration of 3-5 years

• Sharpe ratio of 1.5

• ROE must be greater than cost of equity capital, or, agency

problem

• The answer is probably a little of both

43 Eric Falkenstein 4/14/99

 An Application of BSM VaR

• The average return of short funding a long maturity bond

is positive, as the yield curve is usually upward sloping

• The Sharpe ratio of this risk-return relationship can help

one see the relative attractiveness of this strategy

• With derivatives one can see how to optimize a common,

simply strategy

44 Eric Falkenstein 4/14/99

 Yield Curve

7
RISK
YIELD (%)

PREMIUM

5 EXPECTED
YIELD

4
1 5 10 20
MATURITY
45 Eric Falkenstein 4/14/99
 Returns to Riding the Yield Curve

46 Eric Falkenstein 4/14/99

 1/83-7/98 Performance

Return from being long the specific maturity,

and short a 6 month security, in basis points

1yr 3yr 5yr 10yr 30yr

Avg. Return 36 121 209 300 438

Avg. Volatility 73 161 423 601 876
Sharpe Ratio .49 .75 .49 .50 .50

47 Eric Falkenstein 4/14/99

 Replicate the 5, 10 and 30 Year Strategies
by Replicating the 3 Year Return with a
Zero-cost Derivative

48 Eric Falkenstein 4/14/99

 Applications are Important

• Wholesale replacement of accounting based measures of

risk will not occur soon

• In the meantime, piecemeal applications of VaR to balance

sheet management strategies are quite fruitful

49 Eric Falkenstein 4/14/99

 Diversification Lowers Risk

• Assume portfolio with Sensitivity to 1 Risk Factor,

VaR=100
• Add a portfolio with a sensitivity to a new risk factor
• See new Portfolio VaR by various Correlation and Relative
Size combinations
Correlations
0 0.25 0.5 0.75 1.00
10 101 103 105 108 110
size 50 112 122 132 141 150
100 141 158 173 187 200

50 Eric Falkenstein 4/14/99

 Conclusion

• VaR is applicable across a variety of instruments, and thus

allows a meaningful aggregation. This diminishes the
numbers necessary for a complete picture of risk

• For Trading Desks, VaR is a necessary component of a

regulatory capital charge

• VaR should always be complimented with stress tests

• While VaR is not as applicable to A/L management vis-à-

vis trading desks, piecemeal applications are fruitful

51 Eric Falkenstein 4/14/99